Research On EnhanceIO Implementation Principle And Improvement Scheme

Hybrid storage system solves the problem that the traditional Hard Disk Drive(HDD) based storage system has poor performance and using Solid State Drive(SSD) to replace HDD will cost a lot of money, has a very high application value. EnhanceIO is an open source hybrid storage system software, developed by STEC company and applied to the Linux system. It achieves a block device driver in kernel space to process application data read/write requests, and has a significant improvement on the performance of the storage system.This thesis expounds the key technology of hybrid storage system design, analyzes the prin-ciple of EnhanceIO system in detail, puts forward and achieves an improvement scheme. On the one hand, among the three cache replacement algorithms EnhanceIO system im-plemented, the random(RAND) replacement algorithm completely does not consider the cache data block usage in the history, the first in first out(FIFO) replacement algorithm only considers the order of data blocks to join the cache, the least recently used(LRU) replace-ment algorithm only considers the recent usage of data blocks in the cache. Because there is no comprehensive consideration about the data access time and access frequency of the cache blocks, cannot make a good judgment and prediction about the data access pattern of two storage system, the hit ratio of these three cache replacement algorithms has room for improvement. This thesis, based on the LRU replacement algorithm, using the LRU lists, that is, a low frequency access list and a high frequency access list, achieves a replacement algorithm which comprehensively takes the access time and frequency characteristics into consideration. On the other hand, EnhanceIO system does not distinguish the read/write re-quest type, and caches the read/write requests both in random type and sequential type. The large amount and low frequency sequential type data read/write, such as disk scan and file copy, can cause fluctuations for read/write performance of the storage system. Compared to HDD, SSD has more obvious advantage in the random read/write speed than sequential read/write speed. Distinguishing the type of read/write requests, giving priority to random read/write requests caching, would bring greater performance improvement of the storage system. In order to solve this problem, this thesis joins sequential read/write bypass function in the EnhanceIO system.Through the analysis of test results, compared to the original replacement LRU algorithm, the cache hit ratio of the improved replacement algorithm increased by 3.42%, improving the performance of the storage system. Characterized by respectively, random read test, per- formance increased by 6.94%. Random write test, write performance increased by 4.29%. Random read/write test, read performance increased by 5.52%, write performance increased by 4.48%. In the large amount sequential read/write application scenarios, the addition of sequential read/write bypass function prevents useful data blocks in the cache from being eliminated, making the performance of the storage system more stable.